Internally disable subrange workaround

[CaseyCarter]

The workaround for GH-2326 isn't needed for the internal MSVC compiler; let's ensure it doesn't regress. Since this workaround is in product code let's rename MSVC_INTERNAL_TESTING to _MSVC_INTERNAL_TESTING (because _Ugly). Don't forget to update the name in $/src/qa/distrib.yml consistently when merging.

Resolves #2326

[AlexGuteniev]

Should we disable it based on the compiler full version instead of being internal?

[CaseyCarter]

Should we disable it based on the compiler full version instead of being internal?

I don't trust the full version at all. It doesn't help that I don't know when this bug was fixed other than "between the 17.1p1 snap and last Friday." The goal is to ensure that the compiler FE doesn't regress this fixing something else, for which it suffices to only have coverage when the pertinent test is run in the internal test harness. Yes, that means the workaround won't be disabled outside of the internal harness until we do so manually, and I'm ok with consequently shipping it for an extra cycle.

[StephanTLavavej]

Another reason that we traditionally avoid inspecting _MSC_FULL_VER is that it can vary between different MSVC-internal branches, which historically caused trouble as code got merged here and there. This is probably less of an issue these days where there are only 3 major branches for the current version (prod/fe where compiler front-end and library code goes, prod/be where compiler back-end changes happen, and the main branch where they meet), but the conditioning was long-lasting.

[StephanTLavavej]

⚠️ MSVC-PR mirroring note:

This will require a one-line change to the definition of the macro.

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[StephanTLavavej]

Thanks for ensuring that this compiler bugfix stays fixed, and for cleaning up this macro! (I made the necessary internal change when merging.) 🐞 ✔️ 😺